Commit Graph

113 Commits

Author SHA1 Message Date
Richard van der Hoff
11a67b7c9d
Merge pull request #3093 from matrix-org/rav/response_cache_wrap
Refactor ResponseCache usage
2018-04-20 11:31:17 +01:00
Richard van der Hoff
d3347ad485 Revert "Use sortedcontainers instead of blist"
This reverts commit 9fbe70a7dc.

It turns out that sortedcontainers.SortedDict is not an exact match for
blist.sorteddict; in particular, `popitem()` removes things from the opposite
end of the dict.

This is trivial to fix, but I want to add some unit tests, and potentially some
more thought about it, before we do so.
2018-04-13 11:16:43 +01:00
Richard van der Hoff
60f6014bb7 ResponseCache: fix handling of completed results
Turns out that ObservableDeferred.observe doesn't return a deferred if the
result is already completed. Fix handling and improve documentation.
2018-04-13 07:32:29 +01:00
Richard van der Hoff
b78395b7fe Refactor ResponseCache usage
Adds a `.wrap` method to ResponseCache which wraps up the boilerplate of a
(get, set) pair, and then use it throughout the codebase.

This will be largely non-functional, but does include the following functional
changes:

* federation_server.on_context_state_request: drops use of _server_linearizer
  which looked redundant and could cause incorrect cache misses by yielding
  between the get and the set.
* RoomListHandler.get_remote_public_room_list(): fixes logcontext leaks
* the wrap function includes some logging. I'm hoping this won't be too noisy
  on production.
2018-04-12 13:02:15 +01:00
Richard van der Hoff
d5c74b9f6c
Merge pull request #3092 from matrix-org/rav/response_cache_metrics
Add metrics for ResponseCache
2018-04-12 12:59:36 +01:00
Richard van der Hoff
261124396e
Merge pull request #3059 from matrix-org/rav/doc_response_cache
Document the behaviour of ResponseCache
2018-04-12 11:22:30 +01:00
Richard van der Hoff
b3384232a0 Add metrics for ResponseCache 2018-04-10 23:14:47 +01:00
Vincent Breitmoser
9fbe70a7dc Use sortedcontainers instead of blist
This commit drop-in replaces blist with SortedContainers. They are
written in pure python so work with pypy, but perform as good as
native implementations, at least in a couple benchmarks:

http://www.grantjenks.com/docs/sortedcontainers/performance.html
2018-04-10 11:29:51 +02:00
Richard van der Hoff
01afc563c3 Fix overzealous cache invalidation
Fixes an issue where a cache invalidation would invalidate *all* pending
entries, rather than just the entry that we intended to invalidate.
2018-04-05 16:24:04 +01:00
Richard van der Hoff
a9a74101a4 Document the behaviour of ResponseCache
it looks like everything that uses ResponseCache expects to have to
`make_deferred_yieldable` its results. It's debatable whether that is the best
approach, but let's document it for now to avoid further confusion.
2018-04-04 09:06:22 +01:00
Erik Johnston
9a0d783c11 Add comments 2018-03-19 11:35:53 +00:00
Erik Johnston
7c7706f42b Fix bug where state cache used lots of memory
The state cache bases its size on the sum of the size of entries. The
size of the entry is calculated once on insertion, so it is important
that the size of entries does not change.

The DictionaryCache modified the entries size, which caused the state
cache to incorrectly think it was smaller than it actually was.
2018-03-15 15:46:54 +00:00
Richard van der Hoff
bc496df192 report metrics on number of cache evictions 2018-02-05 15:34:01 +00:00
Erik Johnston
495f075b41 Increase default cache factor size. 2017-07-04 09:58:32 +01:00
Erik Johnston
b5e8d529e6 Define CACHE_SIZE_FACTOR once 2017-07-04 09:56:44 +01:00
Erik Johnston
c72058bcc6 Use an ExpiringCache for storing registration sessions
This is because pruning them was a significant performance drain on
matrix.org
2017-06-29 14:08:37 +01:00
Erik Johnston
efc2b7db95 Rewrite conditional 2017-06-09 13:35:15 +01:00
Erik Johnston
eed59dcc1e Fix has_any_entity_changed
Occaisonally has_any_entity_changed would throw the error: "Set changed
size during iteration" when taking the max of the `sorteddict`. While
its uncertain how that happens, its quite inefficient to iterate over
the entire dict anyway so we change to using the more traditional
`bisect_*` functions.
2017-06-09 11:44:01 +01:00
Erik Johnston
304880d185 Add stream change cache 2017-05-31 15:46:36 +01:00
Erik Johnston
bd7bb5df71 Pull out if statement from for loop 2017-05-22 15:12:19 +01:00
Erik Johnston
e3417a06e2 Update list cache to handle one arg case
We update the normal cache descriptors to handle caches with a single
argument specially so that the key wasn't a 1-tuple. We need to update
the cache list to be aware of this.
2017-05-22 15:04:42 +01:00
Erik Johnston
bbfe4e996c Make get_state_groups_from_groups faster.
Most of the time was spent copying a dict to filter out sentinel values
that indicated that keys did not exist in the dict. The sentinel values
were added to ensure that we cached the non-existence of keys.

By updating DictionaryCache to keep track of which keys were known to
not exist itself we can remove a dictionary copy.
2017-05-17 15:12:15 +01:00
Erik Johnston
ffad4fe35b Don't update event cache hit ratio from get_joined_users
Otherwise the hit ration of plain get_events gets completely skewed by
calls to get_joined_users* functions.
2017-05-08 16:06:17 +01:00
Erik Johnston
d2d8ed4884 Optimise caches with single key 2017-05-04 14:18:46 +01:00
Erik Johnston
efab1dadde Remove DEBUG_CACHES 2017-04-25 10:54:09 +01:00
Erik Johnston
119cb9bbcf Reduce cache size by not storing deferreds
Currently the cache descriptors store deferreds rather than raw values,
this is a simple way of triggering only one database hit and sharing the
result if two callers attempt to get the same value.

However, there are a few caches that simply store a mapping from string
to string (or int). These caches can have a large number of entries,
under the assumption that each entry is small. However, the size of a
deferred (specifically the size of ObservableDeferred) is signigicantly
larger than that of the raw value, 2kb vs 32b.

This PR therefore changes the cache descriptors to store the raw values
rather than the deferreds.

As a side effect cached storage function now either return a deferred or
the actual value, as the cached list decriptor already does. This is
fine as we always end up just yield'ing on the returned value
eventually, which handles that case correctly.
2017-04-25 10:23:11 +01:00
Erik Johnston
d134d0935e Only intern ascii strings 2017-04-24 14:07:48 +01:00
Erik Johnston
4d17add8de Remove unused instance variable 2017-03-31 09:38:27 +01:00
Erik Johnston
6194a64ae9 Doc new instance variables 2017-03-30 14:19:10 +01:00
Erik Johnston
014fee93b3 Manually calculate cache key as getcallargs is expensive
This is because getcallargs recomputes the getargspec, amongst other
things, which we don't need to do as its already been done
2017-03-30 14:14:46 +01:00
Erik Johnston
86780a8bc3 Don't convert to deferreds when not necessary 2017-03-30 14:14:36 +01:00
Richard van der Hoff
f9b4bb05e0 Fix the logcontext handling in the cache wrappers (#2077)
The cache wrappers had a habit of leaking the logcontext into the reactor while
the lookup function was running, and then not restoring it correctly when the
lookup function had completed. It's all the fault of
`preserve_context_over_{fn,deferred}` which are basically a bit broken.
2017-03-30 13:22:24 +01:00
Richard van der Hoff
95f21c7a66 Fix caching of remote servers' signature keys
The `@cached` decorator on `KeyStore._get_server_verify_key` was missing
its `num_args` parameter, which meant that it was returning the wrong key for
any server which had more than one recorded key.

By way of a fix, change the default for `num_args` to be *all* arguments. To
implement that, factor out a common base class for `CacheDescriptor` and `CacheListDescriptor`.
2017-03-22 15:11:30 +00:00
Richard van der Hoff
29ed09e80a Fix assertion to stop transaction queue getting wedged
... and update some docstrings to correctly reflect the types being used.

get_new_device_msgs_for_remote can return a long under some circumstances,
which was being stored in last_device_list_stream_id_by_dest, and was then
upsetting things on the next loop.
2017-03-15 12:16:55 +00:00
Erik Johnston
3545e17f43 Add setdefault key to ExpiringCache 2017-03-10 10:30:49 +00:00
Erik Johnston
6b61060b51 Comment 2017-02-02 14:47:15 +00:00
Erik Johnston
9efcc3f3be Comment 2017-02-02 13:50:22 +00:00
Erik Johnston
c430111d0e Update LruCache size estimate on clear 2017-01-18 14:55:23 +00:00
Erik Johnston
380dba1020 Measure metrics of string_cache 2017-01-17 17:04:46 +00:00
Erik Johnston
37b4c7d8a9 Fix typo in return type 2017-01-17 14:43:32 +00:00
Erik Johnston
d6c75cb7c2 Rename and comment tree_to_leaves_iterator 2017-01-17 11:47:03 +00:00
Erik Johnston
1ccd5676e3 Remove needless call to evict() 2017-01-17 11:42:26 +00:00
Erik Johnston
f85b6ca494 Speed up cache size calculation
Instead of calculating the size of the cache repeatedly, which can take
a long time now that it can use a callback, instead cache the size and
update that on insertion and deletion.

This requires changing the cache descriptors to have two caches, one for
pending deferreds and the other for the actual values. There's no reason
to evict from the pending deferreds as they won't take up any more
memory.
2017-01-17 11:18:13 +00:00
Erik Johnston
6d00213e80 Use OrderedDict in ExpiringCache 2017-01-16 15:33:22 +00:00
Erik Johnston
46aebbbcbf Add support for 'iterable' to ExpiringCache 2017-01-16 14:57:23 +00:00
Erik Johnston
2fae34bd2c Optionally measure size of cache by sum of length of values 2017-01-13 17:46:17 +00:00
Erik Johnston
955f34d23e Change get_pos_of_last_change to return upper bound 2016-09-15 15:12:07 +01:00
Erik Johnston
cb3edec6af Use stream_change cache to make get_forward_extremeties_for_room cache more effective 2016-09-15 14:28:13 +01:00
Erik Johnston
45fd2c8942 Ensure invalidation list does not grow unboundedly 2016-08-19 16:09:16 +01:00
Erik Johnston
c0d7d9d642 Rename to on_invalidate 2016-08-19 15:13:58 +01:00